Domain-wise Invariant Learning for Panoptic Scene Graph Generation
This addresses annotation bias in PSG models, which impedes their real-world utility, though it appears incremental as it builds on existing benchmark models.
The paper tackles biased predicate annotations in Panoptic Scene Graph Generation (PSG) by proposing a framework that infers biased annotations through predicate prediction risks and adaptively transfers them to consistent ones via invariant representation learning. Experiments show the method significantly improves benchmark models, achieving new state-of-the-art performance with demonstrated generalization on PSG datasets.
Panoptic Scene Graph Generation (PSG) involves the detection of objects and the prediction of their corresponding relationships (predicates). However, the presence of biased predicate annotations poses a significant challenge for PSG models, as it hinders their ability to establish a clear decision boundary among different predicates. This issue substantially impedes the practical utility and real-world applicability of PSG models. To address the intrinsic bias above, we propose a novel framework to infer potentially biased annotations by measuring the predicate prediction risks within each subject-object pair (domain), and adaptively transfer the biased annotations to consistent ones by learning invariant predicate representation embeddings. Experiments show that our method significantly improves the performance of benchmark models, achieving a new state-of-the-art performance, and shows great generalization and effectiveness on PSG dataset.